On May 26, 2026, Microsoft pushed out an update that quietly upgrades the AI brain of Windows 11 PCs equipped with Nvidia RTX graphics cards—no driver installation, no user intervention. The update, cataloged as KB5096139, bumps the Nvidia TensorRT-RTX Execution Provider to version 2.2605.1.0 and lands automatically via Windows Update for devices running the specialized Windows 11 26H1 branch.

This isn't a flashy Copilot feature drop or a GeForce game-ready driver. It's a behind-the-scenes component that translates AI model instructions into accelerated workloads on RTX hardware. In plain terms, Microsoft is now treating the AI plumbing between your apps and your GPU as a core Windows service—one it intends to keep current and reliable through the same update channel that delivers security patches and cumulative fixes.

What Actually Changed

KB5096139 replaces an earlier update, KB5089174 (version 2.2604.1.0), and marks the second known servicing of this component through Windows Update. The execution provider it updates sits inside the Windows ML and ONNX Runtime stack, acting as a translator: when an application uses ONNX Runtime to run a machine-learning model locally, the TensorRT-RTX execution provider steps in to compile and run that model on an Nvidia RTX GPU, bypassing the CPU and generic GPU paths for faster inference.

The update page from Microsoft is characteristically terse, noting only “improvements to the execution provider component” for Windows 11 version 26H1. No detailed changelog, no list of optimized operators, no performance benchmarks. But the version bump and the replacement notation confirm this is an iterative release, not a one-off patch. The provider lives under the name “Windows ML Runtime Nvidia TensorRT-RTX Execution Provider Update” in your Update History, and it requires the latest cumulative update for 26H1 to be installed before it will appear.

What This Means for You

For Home Users (Especially New PC Buyers)

If you bought a Windows 11 PC this spring with an RTX GPU and it shipped with version 26H1, you likely already have this update installed. It means that any app using Windows ML or ONNX Runtime for local AI tasks—photo enhancers, video editors, voice processing tools, or upcoming Copilot experiences—can tap your RTX hardware more efficiently. The update does not touch gaming drivers, DirectX, or shader compilers, so your frame rates in Call of Duty won't change. But creative apps like Adobe Photoshop’s neural filters or DaVinci Resolve’s AI features may see subtle speedups or reduced latency.

For the vast majority of users on Windows 11 24H2 or 25H2, this update will not appear. 26H1 is a hardware-optimized release preinstalled on select new devices; it is not offered as an in-place upgrade for existing installations. If you’re not sure which version you have, open Settings > System > About and look for the Windows version number. If it doesn’t read “26H1,” you won’t see KB5096139.

For Developers Building AI-Powered Windows Apps

This update is a quiet but meaningful shift. By servicing the execution provider through Windows Update, Microsoft and Nvidia are telling developers: write your AI code against ONNX Runtime and Windows ML abstractions, and we’ll handle the hardware-specific optimization. That means less time spent bundling and testing TensorRT dependencies, and more time building features.

The trade‑off is that the execution environment may change under your application without notice. An ONNX model that performs perfectly through version 2.2604.1.0 could behave differently after the automatic upgrade to 2.2605.1.0—new memory pressure, different precision handling, or changed initialization times are all possible. Developers targeting this path should include execution provider version in their compatibility matrices and rigorous test cycles, especially for apps that rely on deterministic latency or output stability.

For IT Administrators and Enterprise Fleet Managers

KB5096139 is unlikely to hit your managed devices today, given 26H1’s narrow footprint. But the pattern it establishes is significant. AI acceleration components are now part of the Windows servicing surface, and that means they will eventually enter your update rings, validation groups, and change-management notes.

When 26H2 or a future mainstream build inherits this servicing model, you’ll need to answer new questions: Which AI runtime version is on this PC? Is the execution provider being rolled out in phases? Can we block or delay it with update policies? The old governance mantra of “just check the Windows build number” is starting to fragment. Start thinking about how your inventory tools will surface installed AI components—they’re not going away.

How We Got Here

The path to KB5096139 runs through the past two years of Microsoft’s AI PC strategy. When the company introduced Copilot+ PCs in 2024, it bet on Qualcomm’s NPU for dedicated neural processing. But most desktops and performance laptops still ship with Nvidia GPUs, which possess enormous tensor-processing muscle. Microsoft needed a way to let those devices join the local AI party without requiring every developer to write custom CUDA code.

Enter ONNX Runtime and the execution provider model. ONNX (Open Neural Network Exchange) provides a common format for machine‑learning models; the runtime dispatches those models to whatever accelerator is available. By 2025, Nvidia and Microsoft had crafted the TensorRT‑RTX Execution Provider, a piece of middleware that turns ONNX graphs into optimized RTX inference engines on the fly. It first appeared in developer previews and then in KB5089174, the predecessor to this update.

Windows 11 version 26H1 itself is a peculiar beast. Microsoft announced in February 2026 that 26H1 would not be the next general annual update. Instead, it is a specialized branch for “new hardware,” preloaded on certain devices and not delivered as an upgrade to existing PCs. This split allowed Microsoft to experiment with deeper hardware integration—like serviced AI components—without risking the stability of the much larger 24H2/25H2 install base. KB5096139 fits neatly into that laboratory: a small, silicon‑specific update that may graduate to broader releases once proven.

What to Do Now

If you have a Windows 11 26H1 PC with an RTX GPU:
1. Open Settings > Windows Update > Update history. Look for “Windows ML Runtime Nvidia TensorRT-RTX Execution Provider Update” under the “Other updates” or “Driver updates” section. If you see it, you’re already running version 2.2605.1.0.
2. If you don’t see it, manually check for updates. It should download and install automatically as long as you have the latest cumulative update.
3. No special action is needed. This component sits in the background; it won’t spawn a tray icon or change any settings. Apps will use it when they call ONNX Runtime.

If you’re an app developer:
- Update your test environments to include Windows 11 26H1 with KB5096139 installed.
- Re‑run any ONNX model validation suites to catch regressions early.
- Consider version‑gating your AI features if sensitive to execution provider behavior—while the provider is designed to be drop‑in compatible, AI models can be finicky.

If you’re an IT admin:
- Monitor what devices in your fleet ship with 26H1. If you have pilot units, note whether KB5096139 appears and whether any internal AI‑reliant tools behave differently.
- Start a conversation with line‑of‑business app owners about AI runtime dependencies. The more Microsoft services these components, the more they will resemble critical platform layers like VC++ redistributables or .NET runtimes.

Outlook

KB5096139 is a small update with an awkward name, but it’s a clear signal. Microsoft is knitting local AI acceleration into the operating system’s monthly maintenance fabric. For RTX owners, that means AI features should get faster over time without any tinkering. For the broader Windows ecosystem, it previews a future where your PC’s neural muscle is updated and protected as routinely as its defenses against malware.

Expect similar execution provider updates for other hardware targets—Qualcomm’s NPU, Intel’s AI engine, AMD’s Ryzen AI—to show up in Windows Update as soon as their hosting branches reach maturity. The next phase of the AI PC won’t be announced with a keynote; it will arrive as a couple of megabytes in your nightly update cycle.